Skip to content

EXTERNAL_REVIEW_OPTIONS.md

Latest commit

 

History

History
145 lines (91 loc) · 5.54 KB

EXTERNAL_REVIEW_OPTIONS.md

File metadata and controls

145 lines (91 loc) · 5.54 KB
Raw

External Review Options

Language: English | 中文说明见下方

Core Rule

External review is optional and advisory. It is not evidence, not source support, not compliance approval, not a grade, and not a delivery pass.

You do not need Claude Code to use this starter kit. The default quality path is still local:

  1. source-first checks;
  2. cognitive protocol;
  3. academic self-review loop;
  4. integrity preflight;
  5. document-quality and delivery gates.

External review is an extra second opinion after those local checks.

Three Review Paths

Path Requires Best For Boundary
Local self-review Codex and local project files Default workflow for every user Runs inside your project rules; still needs source checks
Manual external-review bundle Any separate Codex, ChatGPT, Claude, Gemini, or human reviewer Users without Claude Code or users who want a visible copy/paste workflow Bundle is local until you choose to paste it elsewhere
Claude Code runner Claude Code installed and available Users who already have Claude Code and want a direct wrapper Advisory only; not required

Path 1: Local Self-Review

Use the built-in self-review workflow first. It checks writing quality and revision priorities without sending the artifact to another service.

For formal research writing, the project should route through:

  • cognitive-frameworks;
  • academic-self-review-loop;
  • academic-integrity-preflight;
  • dissertation-document-quality-gate.

Path 2: Manual Review Bundle For Codex / ChatGPT / Claude / Gemini

Build a local review bundle:

python3 scripts/build_external_review_bundle.py path/to/draft.md

The script creates a folder under .agent-runtime/external-review-bundles/.

Open the generated folder and inspect:

  • privacy_scan.md;
  • artifact.md;
  • EXTERNAL_REVIEW_PROMPT.md;
  • manifest.json.

If the privacy scan is acceptable, copy EXTERNAL_REVIEW_PROMPT.md into a separate Codex window, ChatGPT chat, Claude chat, Gemini chat, or give it to a human reviewer.

The bundle builder does not call an LLM, does not connect to the internet, and does not upload files.

Path 3: Claude Code Runner

If Claude Code is installed and available, you can run:

python3 scripts/claude_independent_review.py path/to/draft.md

This is one optional runner for the same external-review role. Claude feedback is advisory and must become a revision queue. It does not replace local source, privacy, citation, compliance, or delivery gates.

Sensitive Material

The bundle builder and Claude runner block common sensitive content by default, including participant-related text, interview material, consent language, emails, private paths, tokens, and credentials.

If a sensitive item is blocked, use an anonymised draft. Only use override flags when you explicitly accept the risk.

中文

核心规则

External review 是可选的建议层。它不是证据、不是 source support、不是合规批准、不是分数，也不是正式交付通过。

这套 starter kit 不需要 Claude Code。默认质量流程仍然是本地流程：

  1. source-first checks；
  2. cognitive protocol；
  3. academic self-review loop；
  4. integrity preflight；
  5. document-quality and delivery gates。

External review 只是这些本地检查之后的额外第二意见。

三种质审路径

路径 需要什么 适合什么情况 边界
本地 self-review Codex 和本地项目文件 所有用户的默认流程 在本地项目规则内运行，但仍需要 source checks
手动 external-review bundle 另一个 Codex、ChatGPT、Claude、Gemini 或人工 reviewer 没有 Claude Code 的用户，或想要可见复制粘贴流程的用户 bundle 只在本地生成，除非你主动复制出去
Claude Code runner 已安装并可用的 Claude Code 已有 Claude Code 的用户 只提供建议，不是必需工具

路径一：本地 Self-Review

先使用内置 self-review workflow。它可以在不发送给外部服务的情况下检查写作质量和修改优先级。

正式研究写作通常应经过：

  • cognitive-frameworks；
  • academic-self-review-loop；
  • academic-integrity-preflight；
  • dissertation-document-quality-gate。

路径二：给 Codex / ChatGPT / Claude / Gemini 的手动 Review Bundle

生成本地质审包：

python3 scripts/build_external_review_bundle.py path/to/draft.md

脚本会在 .agent-runtime/external-review-bundles/ 下创建一个文件夹。

打开生成的文件夹，检查：

  • privacy_scan.md；
  • artifact.md；
  • EXTERNAL_REVIEW_PROMPT.md；
  • manifest.json。

如果隐私扫描可以接受，把 EXTERNAL_REVIEW_PROMPT.md 复制到另一个 Codex 窗口、ChatGPT、Claude、Gemini，或交给人工 reviewer。

这个 bundle builder 不调用 LLM，不联网，也不上传文件。

路径三：Claude Code Runner

如果你已经安装并能使用 Claude Code，可以运行：

python3 scripts/claude_independent_review.py path/to/draft.md

这只是同一个 external-review 角色的一个可选 runner。Claude feedback 只能作为 revision queue，不能替代本地 source、privacy、citation、compliance 或 delivery gates。

敏感材料

Bundle builder 和 Claude runner 默认会拦截常见敏感内容，包括 participant 相关文本、interview material、consent language、emails、private paths、tokens 和 credentials。

如果内容被拦截，优先使用匿名化草稿。只有在你明确接受风险时，才使用 override flags。